Anthropic, a research institute known for its focus on safe AI development, has recruited Jan Leike, a prominent figure in AI safety. Leike previously led OpenAI’s “superalignment” team, dedicated to ensuring advanced AI remains aligned with human values.
Leike’s departure from OpenAI reportedly stemmed from disagreements over the organization’s direction, with Leike advocating for a stronger safety focus. Now at Anthropic, he will head up a newly formed “superalignment” team, mirroring his prior role. This team will concentrate on critical areas of AI safety, including scalable oversight methods, ensuring AI systems remain controllable as they grow in complexity.
Anthropic has consistently positioned itself as a leader in responsible AI development, often contrasting its approach with OpenAI’s perceived prioritization of commercial pursuits. The appointment of Leike, a vocal proponent of robust safety measures, reinforces this stance.
Leike’s team will delve into areas like “weak-to-strong generalization,” aiming to ensure AI systems perform as intended in real-world situations that may differ from their training data. Additionally, they will explore “automated alignment research,” a field seeking to develop mechanisms for automatically aligning AI goals with human values.
While OpenAI has yet to comment on Leike’s departure, Anthropic CEO Dario Amodei, who himself left OpenAI over disagreements about direction, expressed his enthusiasm for Leike’s leadership. Amodei emphasized Anthropic’s commitment to tackling the challenges of superintelligence and ensuring AI safety remains paramount.
This recruitment has significant implications for the future of AI development. The competition between Anthropic and OpenAI could accelerate advancements in AI safety research as both institutions strive for leadership in this crucial domain. However, concerns linger about the potential talent drain at OpenAI and the impact on its own safety efforts.
